Vercel's Chat SDK deploys a Claude Managed Agent as a Slack bot
No new frontier model tonight, but the plumbing moved: one-key AI Gateway routes, Cursor as an AI SDK harness, and Granite 4.2 open weights to run local.
Copy markdown
Deploy a Claude agent as a Slack bot
Vercel's Chat SDK ships a template that stands up a Claude Managed Agent as a Slack research bot, with persistent sessions per thread and streamed briefs that cite their sources. It's a production agent with server-side memory you deploy — no queue, session store, or streaming layer to build yourself.
Eve agents, built from the dashboard
You can now scaffold, Git-connect, and deploy an 'eve' agent straight from the Vercel dashboard, pre-wired to AI Gateway models and MCP connections. Zero to a running agent without leaving the browser or touching a terminal.
Cursor becomes an AI SDK harness
The new @ai-sdk/harness-cursor adapter exposes Cursor through the same HarnessAgent interface as every other harness, so you can drive Cursor agents programmatically next to Claude Code and Codex in one codebase. Useful for evals or multi-agent orchestration across tools.
Meta's Muse Image, one key away
Meta's Muse Image is now callable through Vercel AI Gateway with a single key — automatic fallbacks, spend tracking, and request traces, and no Meta provider account to set up. A hosted image model on the same gateway as your text models.
Granite 4.2: a reasoning agent you run local
IBM's Apache-2.0 Granite 4.2 opens at 3B/8B/30B with a 'thinking switch,' native tool calling, and a 128K context (512K extendable). The 30B scores 57.0 on SWE-bench Verified and 61.4 on BFCL v4 — a genuinely agentic small model on Hugging Face and Ollama today.